AI Fundamentals for Instructional Designers

AI Fundamentals for Instructional Designers

To design effective learning experiences with AI, we must first understand the "engine" that powers our tools. This chapter demystifies Large Language Models (LLMs) and explains the core concepts that every instructional designer should know.

1. What is an LLM?

A Large Language Model is a type of artificial intelligence trained on massive amounts of text data. It uses statistical patterns to predict the next word (or "token") in a sequence.

[!NOTE] Think of an LLM as a highly sophisticated "auto-complete" built on the sum of human digital knowledge.

For an ID, an LLM is more than a chatbot; it is a reasoning engine. It can synthesize information, take on personas (e.g., "Act as a subject matter expert in physics"), and format content into specific structures (e.g., "Generate a SCORM-compliant outline").

2. Tokens and Context Windows

Understanding how AI "reads" and "remembers" is crucial for prompt engineering.

Tokens

AI doesn't read words like humans do. It breaks text into tokens—small chunks of characters. - Rule of Thumb: 1,000 tokens ≈ 750 words. - Visual: A standard page of single-spaced text is about 500 words, or ~660 tokens.

Why does this matter? API costs and model limits are often based on token counts.

Context Window

The context window is the amount of information the model can "hold in its head" at once during a conversation. In 2025, context windows have expanded significantly (with some models handling millions of tokens), but the core principle remains: the more relevant context you provide in your prompt, the better the output.

3. Generative vs. Discriminative AI

  • Generative AI: Creates new content (text, images, video) based on patterns. This is where most ID work happens (e.g., creating case studies).
  • Discriminative AI: Classifies or analyzes existing data. In ID, this is used for grading, identifying gaps in a curriculum, or sentiment analysis of learner feedback.

4. The AI Toolbox: Beyond ChatGPT

While ChatGPT (and other LLMs like Claude or Gemini) are the most famous tools, the AI landscape for instructional designers is vast. As recent reviews by The eLearning Coach and Cathy Moore highlight, the toolkit can be categorized by function:

  • Writing & Content: Jasper, ChatGPT, Claude (for brainstorming, drafting scenarios, and rewriting content).
  • Multimedia Generation:
    • Video: Synthesia (AI avatars), HeyGen.
    • Audio: WellSaid Labs (high-fidelity voiceovers), Descript (audio/video editing with text).
    • Presentation: Tome (rapid slide generation).
  • Research & Synthesis: Elicit, Consensus (finding academic papers and summarizing research without hallucinations).
  • Productivity: Otter.ai (meeting transcription), Notion AI.

[!TIP] Tools evolve rapidly. Focus on the category of the tool (e.g., "AI Voice Generator") rather than becoming dependent on a single brand.

5. Evaluating AI Tools: A Framework for IDs

With hundreds of new AI tools launching every week, how do you choose the right one? Use this simple checklist before adopting a new tool in your workflow:

Criteria Key Question
Privacy & Security Does this tool use my data to train its public models? (If yes, do NOT use for proprietary content).
Accuracy (Hallucination) How does the tool cite its sources? Can I verify the output easily?
Cost vs. ROI Does the time saved by this tool justify the subscription cost? (e.g., A $30/mo video generator is worth it if it saves 10 hours of animation work).
Exportability Can I easily export the content to my LMS or authoring tool (e.g., SCORM, HTML5, MP4)?
Accessibility Does the output meet WCAG standards (e.g., auto-captions for video)?

[!TIP] Start Small: Don't try to overhaul your entire process at once. Pick one tool to solve one specific bottleneck (e.g., "I need faster audio narration") and evaluate its ROI for that specific task.

6. The "Hallucination" Problem

One of the biggest challenges in AI-Powered ID is hallucination—when the model generates factually incorrect information that sounds highly convincing.

[!IMPORTANT] Never use AI-generated content in a learning module without a Subject Matter Expert (SME) or your own rigorous verification. AI is a creative assistant, not an encyclopedia.

7. Ethical Considerations

As instructional designers, we have a responsibility to our learners:

  • Bias: AI models can inherit biases from their training data. We must audit outputs for gender, racial, and cultural bias.
  • Privacy: Never input sensitive student data or proprietary company information into public AI models.
  • Accessibility: Ensure AI-generated content (images, video, text) meets WCAG 2.1 standards. AI can help generate alt-text, but a human must verify it for accuracy and context.
  • Academic Integrity: We must design assessments that focus on higher-order thinking (Bloom’s Taxonomy) which AI cannot easily replicate without human synthesis.

Reflection Exercise

  1. Choose a topic you are currently teaching.
  2. Write a prompt for an LLM to generate three different ways to introduce that topic.
  3. Evaluate the output: Which one is the most engaging? Did the model hallucinate any facts?

References:

  • EDUCAUSE (2025). 2025 Horizon Report: Teaching and Learning Edition.
  • Malamed, C. (2025). AI Tools for Instructional Designers. The eLearning Coach.
  • Mollick, E. (2024). Co-Intelligence: Living and Working with AI. Portfolio.
  • Moore, C. (2025). Best AI Tools for Instructional Designers.